Tesorito

AI-powered financial agent that gives Peruvian micro-SMEs real-time cash flow visibility and automatic bank reconciliation, through a simple chat.

Comment

Inspiration

Micro and small businesses run a huge part of Peru's economy, but almost none of them have real visibility into their own cash flow. Most financial tools assume you already have an accountant and clean digital records, an assumption that just doesn't hold for a corner store or a small family business juggling receipts and three different bank apps. We wanted something that felt less like a dashboard and more like a conversation, a tool small business owners would actually open every day.

What it does

Tesorito is an AI-powered financial agent for Peruvian micro-SMEs. Through an embedded chat, it automatically reconciles bank transactions, predicts cash flow, and surfaces alerts before problems become emergencies. A live demo scheduler lets anyone try the product hands-on instead of just watching a slide deck.

How we built it

We built Tesorito as a Next.js/TypeScript frontend on top of a FastAPI backend, using LangGraph to orchestrate AI reasoning and Supabase for data persistence. An n8n orchestrator runs three separate workflows behind the scenes, and a FastAPI bank simulator covers three Peruvian banks with 130 historical transactions so we could test reconciliation logic realistically. Claude (Sonnet 4.6) powers entity extraction and alerting inside the chat.

Challenges we ran into

Bank reconciliation was by far the hardest part. Each bank's statement format is messy and inconsistent in its own way, and matching transactions across all three without hardcoding every edge case almost broke us the night before demos. Getting a reconciliation approach that generalized, instead of just working for our test data, taught us more about building resilient data pipelines than any single class had.

Accomplishments that we're proud of

We're proud that Tesorito went from an idea born at IDEATECH PYME 2026 to a working product with real bank simulation and predictive cash flow, all within a hackathon timeline, and that it won first place.

What we learned

We learned how much design work goes into making an AI product feel trustworthy for a non-technical user, and how much messier real financial data is than any sample dataset suggests.

What's next for Tesorito

We're actively developing the bank reconciliation module further and generating SIRE API reference documentation (covering 30 RCE and 16 RVIE endpoints) to support Tesorito's SUNAT integrations, moving toward real bank and tax authority connectivity.

Built With

Share this project:

Updates